A1Mobilsoft's Operations Inquiry Automation Journey (1): Problem Definition and 3-Agent Architecture Design

A1Mobilsoft introduced a Multi-Agent system to improve response efficiency for a Slack channel that saw 846 operations inquiries over 32 months. The existing manual response method, which took an average of 2.6 hours, was causing problems with developer work engagement.

In particular, three major patterns caused operational pain: repetitive problem investigation, complex repository identification, and response delays when the responsible person was unavailable. To solve this, they designed an architecture utilizing Amazon Bedrock AgentCore Runtime, Bedrock Knowledge Base, and the Strands Agents SDK.

The new system consists of agents with the following three roles to efficiently solve problems.

• Knowledge Agent: Explores past resolution cases and knowledge
• Code Agent: Analyzes and identifies relevant source code
• DB Agent: Checks the actual situation through the database

This project was carried out in collaboration with Megazone Cloud through AWS DevCraft 2.0, and focused on solving the 'tail' problem of drawn-out responses, beyond simply reducing the average time.